Ssd.pytorch: AssertionError: Must define a window to update? why?

[237014845]

iter 2360 || Loss: 9.8369 || timer: 0.5007 sec.
iter 2370 || Loss: 9.9223 || timer: 0.4940 sec.
iter 2380 || Loss: 11.2484 || timer: 0.4787 sec.
iter 2390 || Loss: 10.6926 || timer: 0.4976 sec.
iter 2400 || Loss: 9.6664 || timer: 0.4990 sec.
iter 2410 || Loss: 10.5185 || timer: 0.5092 sec.
iter 2420 || Loss: 12.2446 || timer: 0.5129 sec.
iter 2430 || Loss: 12.2896 || timer: 0.4818 sec.
iter 2440 || Loss: 10.2730 || timer: 0.4859 sec.
iter 2450 || Loss: 10.4930 || timer: 0.5135 sec.
iter 2460 || Loss: 9.9998 || timer: 0.5068 sec.
iter 2470 || Loss: 11.2983 || timer: 0.5139 sec.
iter 2480 || Loss: 10.9896 || timer: 0.4783 sec.
iter 2490 || Loss: 9.8864 || timer: 0.5100 sec.
iter 2500 || Loss: 10.8051 || timer: 0.4982 sec.
iter 2510 || Loss: 10.4417 || timer: 0.4783 sec.
iter 2520 || Loss: 10.2917 || timer: 0.4758 sec.
iter 2530 || Loss: 10.9877 || timer: 0.4810 sec.
iter 2540 || Loss: 10.6649 || timer: 0.4757 sec.
iter 2550 || Loss: 10.9075 || timer: 0.4899 sec.
iter 2560 || Loss: 11.8318 || Traceback (most recent call last):
File "train.py", line 266, in
train()
File "train.py", line 159, in train
'append', epoch_size)
File "train.py", line 261, in update_vis_plot
update=True
File "/home/hs/anaconda3/lib/python3.6/site-packages/visdom/__init__.py", line 288, in wrapped_f
return f(args, *kwargs)
File "/home/hs/anaconda3/lib/python3.6/site-packages/visdom/__init__.py", line 1207, in line
update=update, name=name)
File "/home/hs/anaconda3/lib/python3.6/site-packages/visdom/__init__.py", line 288, in wrapped_f
return f(args, *kwargs)
File "/home/hs/anaconda3/lib/python3.6/site-packages/visdom/__init__.py", line 1010, in scatter
assert win is not None, 'Must define a window to update'
AssertionError: Must define a window to update

[237014845]

train on COCO dataset

[ankitksharma]

I'm getting the error on COCO dataset also. Any solution?

[237014845]

line 163 update_vis_plot(epoch, loc_loss, conf_loss, iter_plot, epoch_plot, 'append', epoch_size)

[liuchang8am]

same error when training on VOC2007

File "train.py", line 257, in <module> train() File "train.py", line 156, in train 'append', viz, epoch_size) File "train.py", line 252, in update_vis_plot update=True File "/home/lc/anaconda3/envs/dl/lib/python3.5/site-packages/visdom/__init__.py", line 288, in wrapped_f return f(*args, **kwargs) File "/home/lc/anaconda3/envs/dl/lib/python3.5/site-packages/visdom/__init__.py", line 1207, in line update=update, name=name) File "/home/lc/anaconda3/envs/dl/lib/python3.5/site-packages/visdom/__init__.py", line 288, in wrapped_f return f(*args, **kwargs) File "/home/lc/anaconda3/envs/dl/lib/python3.5/site-packages/visdom/__init__.py", line 1010, in scatter assert win is not None, 'Must define a window to update' AssertionError: Must define a window to update

[d-li14]

@237014845 hi, you solved the problem by modifying this line: update_vis_plot(epoch, loc_loss, conf_loss, iter_plot, epoch_plot, 'append', epoch_size)? In my opinion, the update_vis_plot function keeps updating parameter win1, i.e. iter_plot instead of the expected epoch_plot.
And I think through just moving the line epoch += 1 before update_vis_plot, this problem can be solved.

[chanajianyu]

anyone solve the problem?

[salt-fisher]

@chanajianyu I have faced the same problem, and adjust the code as your point. It works, thx

[chensonglu]

@d-li14 According to your fix, it works, thx.

[BerryLJ]

iter 730 || Loss: 6.8846 || timer: 0.4548 sec.
iter 740 || Loss: 6.6700 || timer: 0.4608 sec.
iter 750 || Loss: 6.8756 || timer: 0.4498 sec.
iter 760 || Loss: 6.4595 || Traceback (most recent call last):
File "E:/ownstudy/ssd.pytorch-master/train.py", line 272, in
train()
File "E:/ownstudy/ssd.pytorch-master/train.py", line 164, in train
update_vis_plot(epoch, loc_loss, conf_loss, epoch_plot, None ,'append', epoch_size)
File "E:/ownstudy/ssd.pytorch-master/train.py", line 267, in update_vis_plot
update=update_type
File "C:\Users\Berry\Anaconda3\envs\pytorch\lib\site-packages\visdom__init__.py", line 335, in wrapped_f
return f(args, *kwargs)
File "C:\Users\Berry\Anaconda3\envs\pytorch\lib\site-packages\visdom__init__.py", line 1367, in line
update=update, name=name)
File "C:\Users\Berry\Anaconda3\envs\pytorch\lib\site-packages\visdom__init__.py", line 335, in wrapped_f
return f(args, *kwargs)
File "C:\Users\Berry\Anaconda3\envs\pytorch\lib\site-packages\visdom__init__.py", line 1144, in scatter
assert win is not None, 'Must define a window to update'
AssertionError: Must define a window to update

Process finished with exit code 1

Hello, does anyone know the reason? Thank you！

[SalahAdDin]

i'm getting the same error with COCO set.

[SalahAdDin]

@d-li14
https://github.com/d-li14/ssd.pytorch/commit/36fc354a0aed372d1f835efa7ad5684c5ee13daf It is working for me, at least until now.
What's about making a PR?

[Monster-Zhang]

the num_class in config.py should +1 and in ssd.py are not

[shirui-japina]

@237014845 hi, you solved the problem by modifying this line: update_vis_plot(epoch, loc_loss, conf_loss, iter_plot, epoch_plot, 'append', epoch_size)? In my opinion, the update_vis_plot function keeps updating parameter win1, i.e. iter_plot instead of the expected epoch_plot.
And I think through just moving the line epoch += 1 before update_vis_plot, this problem can be solved.

It works,

if args.visdom and iteration != 0 and (iteration % epoch_size == 0):
    epoch += 1
    update_vis_plot(epoch, loc_loss, conf_loss, epoch_plot, None, 'append', epoch_size)

gooood!!